encode: fix commented table with comment (#894)

And updated README.

.github/workflows/coverage.yml

@@ -13,8 +13,8 @@ jobs:
         with:
           fetch-depth: 0
       - name: Setup go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v4
         with:
-          go-version: "1.20"
+          go-version: "1.21"
       - name: Run tests with coverage
         run: ./ci.sh coverage -d "${GITHUB_BASE_REF-HEAD}"

.github/workflows/release.yml

@@ -20,9 +20,9 @@ jobs:
         with:
           fetch-depth: 0
       - name: Set up Go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v4
         with:
-          go-version: "1.20"
+          go-version: "1.21"
       - name: Login to GitHub Container Registry
         uses: docker/login-action@v2
         with:

.github/workflows/workflow.yml

@@ -11,8 +11,8 @@ jobs:
   build:
     strategy:
       matrix:
-        os: [ 'ubuntu-latest', 'windows-latest', 'macos-latest']
+        os: [ 'ubuntu-latest', 'windows-latest', 'macos-latest' ]
-        go: [ '1.19', '1.20' ]
+        go: [ '1.20', '1.21' ]
     runs-on: ${{ matrix.os }}
     name: ${{ matrix.go }}/${{ matrix.os }}
     steps:
@@ -20,7 +20,7 @@ jobs:
         with:
           fetch-depth: 0
       - name: Setup go ${{ matrix.go }}
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v4
         with:
           go-version: ${{ matrix.go }}
       - name: Run unit tests

.goreleaser.yaml

@@ -18,6 +18,7 @@ builds:
       - linux_amd64
       - linux_arm64
       - linux_arm
+      - linux_riscv64
       - windows_amd64
       - windows_arm64
       - windows_arm
@@ -37,6 +38,7 @@ builds:
       - linux_amd64
       - linux_arm64
       - linux_arm
+      - linux_riscv64
       - windows_amd64
       - windows_arm64
       - windows_arm
@@ -55,6 +57,7 @@ builds:
     targets:
       - linux_amd64
       - linux_arm64
+      - linux_riscv64
       - linux_arm
       - windows_amd64
       - windows_arm64

LICENSE

@@ -1,6 +1,7 @@
 The MIT License (MIT)
 
-Copyright (c) 2013 - 2022 Thomas Pelletier, Eric Anderton
+go-toml v2
+Copyright (c) 2021 - 2023 Thomas Pelletier
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -45,16 +45,15 @@ to check for typos. [See example in the documentation][strict].
 
 ### Contextualized errors
 
-When most decoding errors occur, go-toml returns [`DecodeError`][decode-err]),
+When most decoding errors occur, go-toml returns [`DecodeError`][decode-err],
 which contains a human readable contextualized version of the error. For
 example:
 
 ```
-2| key1 = "value1"
+1| [server]
-3| key2 = "missing2"
+2| path = 100
- | ~~~~ missing field
+ |        ~~~ cannot decode TOML integer into struct field toml_test.Server.Path of type string
-4| key3 = "missing3"
+3| port = 50
-5| key4 = "value4"
 ```
 
 [decode-err]: https://pkg.go.dev/github.com/pelletier/go-toml/v2#DecodeError
@@ -73,6 +72,26 @@ representation.
 [tlt]: https://pkg.go.dev/github.com/pelletier/go-toml/v2#LocalTime
 [tldt]: https://pkg.go.dev/github.com/pelletier/go-toml/v2#LocalDateTime
 
+### Commented config
+
+Since TOML is often used for configuration files, go-toml can emit documents
+annotated with [comments and commented-out values][comments-example]. For
+example, it can generate the following file:
+
+```toml
+# Host IP to connect to.
+host = '127.0.0.1'
+# Port of the remote server.
+port = 4242
+
+# Encryption parameters (optional)
+# [TLS]
+# cipher = 'AEAD-AES128-GCM-SHA256'
+# version = 'TLS 1.3'
+```
+
+[comments-example]: https://pkg.go.dev/github.com/pelletier/go-toml/v2#example-Marshal-Commented
+
 ## Getting started
 
 Given the following struct, let's see how to read it and write it as TOML:
@@ -497,27 +516,20 @@ is not necessary anymore.
 
 V1 used to provide multiple struct tags: `comment`, `commented`, `multiline`,
 `toml`, and `omitempty`. To behave more like the standard library, v2 has merged
-`toml`, `multiline`, and `omitempty`. For example:
+`toml`, `multiline`, `commented`, and `omitempty`. For example:
 
 ```go
 type doc struct {
 	// v1
-	F string `toml:"field" multiline:"true" omitempty:"true"`
+	F string `toml:"field" multiline:"true" omitempty:"true" commented:"true"`
 	// v2
-	F string `toml:"field,multiline,omitempty"`
+	F string `toml:"field,multiline,omitempty,commented"`
 }
 ```
 
 Has a result, the `Encoder.SetTag*` methods have been removed, as there is just
 one tag now.
 
-
-#### `commented` tag has been removed
-
-There is no replacement for the `commented` tag. This feature would be better
-suited in a proper document model for go-toml v2, which has been [cut from
-scope][nodoc] at the moment.
-
 #### `Encoder.ArraysWithOneElementPerLine` has been renamed
 
 The new name is `Encoder.SetArraysMultiline`. The behavior should be the same.

ci.sh

@@ -77,8 +77,9 @@ cover() {
 
     pushd "$dir"
     go test -covermode=atomic  -coverpkg=./... -coverprofile=coverage.out.tmp ./...
-    cat coverage.out.tmp | grep -v testsuite | grep -v tomltestgen | grep -v gotoml-test-decoder > coverage.out
+    cat coverage.out.tmp | grep -v fuzz | grep -v testsuite | grep -v tomltestgen | grep -v gotoml-test-decoder > coverage.out
     go tool cover -func=coverage.out
+    echo "Coverage profile for ${branch}: ${dir}/coverage.out" >&2
     popd
 
     if [ "${branch}" != "HEAD" ]; then

decode.go

@@ -318,7 +318,7 @@ func parseFloat(b []byte) (float64, error) {
 	if cleaned[0] == '+' || cleaned[0] == '-' {
 		start = 1
 	}
-	if cleaned[start] == '0' && isDigit(cleaned[start+1]) {
+	if cleaned[start] == '0' && len(cleaned) > start+1 && isDigit(cleaned[start+1]) {
 		return 0, unstable.NewParserError(b, "float integer part cannot have leading zeroes")
 	}
 

example_text_marshaling_test.go

@@ -0,0 +1,37 @@
+package toml_test
+
+import (
+	"fmt"
+	"log"
+	"strconv"
+
+	"github.com/pelletier/go-toml/v2"
+)
+
+type customInt int
+
+func (i *customInt) UnmarshalText(b []byte) error {
+	x, err := strconv.ParseInt(string(b), 10, 32)
+	if err != nil {
+		return err
+	}
+	*i = customInt(x * 100)
+	return nil
+}
+
+type doc struct {
+	Value customInt
+}
+
+func ExampleUnmarshal_textUnmarshal() {
+	var x doc
+
+	data := []byte(`value  = "42"`)
+	err := toml.Unmarshal(data, &x)
+	if err != nil {
+		log.Fatal(err)
+	}
+	fmt.Println(x)
+	// Output:
+	// {4200}
+}

fuzz_test.go

@@ -1,5 +1,5 @@
-//go:build go1.18 || go1.19 || go1.20
+//go:build go1.18 || go1.19 || go1.20 || go1.21
-// +build go1.18 go1.19 go1.20
+// +build go1.18 go1.19 go1.20 go1.21
 
 package toml_test
 

go.mod

@@ -2,4 +2,4 @@ module github.com/pelletier/go-toml/v2
 
 go 1.16
 
-require github.com/stretchr/testify v1.8.1
+require github.com/stretchr/testify v1.8.4

go.sum

@@ -8,8 +8,8 @@ github.com/stretchr/objx v0.4.0/go.mod h1:YvHI0jy2hoMjB+UWwv71VJQ9isScKT/TqJzVSS
 github.com/stretchr/objx v0.5.0/go.mod h1:Yh+to48EsGEfYuaHDzXPcE3xhTkx73EhmCGUpEOglKo=
 github.com/stretchr/testify v1.7.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
 github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU=
-github.com/stretchr/testify v1.8.1 h1:w7B6lhMri9wdJUVmEZPGGhZzrYTPvgJArz7wNPgYKsk=
+github.com/stretchr/testify v1.8.4 h1:CcVxjf3Q8PM0mHUKJCdn+eZZtm5yQwehR5yeSVQQcUk=
-github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4=
+github.com/stretchr/testify v1.8.4/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=

marshaler.go

@@ -148,6 +148,9 @@ func (enc *Encoder) SetIndentTables(indent bool) *Encoder {
 //
 // The "omitempty" option prevents empty values or groups from being emitted.
 //
+// The "commented" option prefixes the value and all its children with a comment
+// symbol.
+//
 // In addition to the "toml" tag struct tag, a "comment" tag can be used to emit
 // a TOML comment before the value being annotated. Comments are ignored inside
 // inline tables. For array tables, the comment is only present before the first
@@ -180,6 +183,7 @@ func (enc *Encoder) Encode(v interface{}) error {
 type valueOptions struct {
 	multiline bool
 	omitempty bool
+	commented bool
 	comment   string
 }
 
@@ -205,6 +209,9 @@ type encoderCtx struct {
 	// Indentation level
 	indent int
 
+	// Prefix the current value with a comment.
+	commented bool
+
 	// Options coming from struct tags
 	options valueOptions
 }
@@ -273,7 +280,7 @@ func (enc *Encoder) encode(b []byte, ctx encoderCtx, v reflect.Value) ([]byte, e
 		return enc.encodeMap(b, ctx, v)
 	case reflect.Struct:
 		return enc.encodeStruct(b, ctx, v)
-	case reflect.Slice:
+	case reflect.Slice, reflect.Array:
 		return enc.encodeSlice(b, ctx, v)
 	case reflect.Interface:
 		if v.IsNil() {
@@ -357,6 +364,7 @@ func (enc *Encoder) encodeKv(b []byte, ctx encoderCtx, options valueOptions, v r
 
 	if !ctx.inline {
 		b = enc.encodeComment(ctx.indent, options.comment, b)
+		b = enc.commented(ctx.commented, b)
 		b = enc.indent(ctx.indent, b)
 	}
 
@@ -378,6 +386,13 @@ func (enc *Encoder) encodeKv(b []byte, ctx encoderCtx, options valueOptions, v r
 	return b, nil
 }
 
+func (enc *Encoder) commented(commented bool, b []byte) []byte {
+	if commented {
+		return append(b, "# "...)
+	}
+	return b
+}
+
 func isEmptyValue(v reflect.Value) bool {
 	switch v.Kind() {
 	case reflect.Struct:
@@ -526,6 +541,8 @@ func (enc *Encoder) encodeTableHeader(ctx encoderCtx, b []byte) ([]byte, error)
 
 	b = enc.encodeComment(ctx.indent, ctx.options.comment, b)
 
+	b = enc.commented(ctx.commented, b)
+
 	b = enc.indent(ctx.indent, b)
 
 	b = append(b, '[')
@@ -577,11 +594,23 @@ func (enc *Encoder) encodeKey(b []byte, k string) []byte {
 	}
 }
 
-func (enc *Encoder) encodeMap(b []byte, ctx encoderCtx, v reflect.Value) ([]byte, error) {
+func (enc *Encoder) keyToString(k reflect.Value) (string, error) {
-	if v.Type().Key().Kind() != reflect.String {
+	keyType := k.Type()
-		return nil, fmt.Errorf("toml: type %s is not supported as a map key", v.Type().Key().Kind())
+	switch {
-	}
+	case keyType.Kind() == reflect.String:
+		return k.String(), nil
 
+	case keyType.Implements(textMarshalerType):
+		keyB, err := k.Interface().(encoding.TextMarshaler).MarshalText()
+		if err != nil {
+			return "", fmt.Errorf("toml: error marshalling key %v from text: %w", k, err)
+		}
+		return string(keyB), nil
+	}
+	return "", fmt.Errorf("toml: type %s is not supported as a map key", keyType.Kind())
+}
+
+func (enc *Encoder) encodeMap(b []byte, ctx encoderCtx, v reflect.Value) ([]byte, error) {
 	var (
 		t                 table
 		emptyValueOptions valueOptions
@@ -589,13 +618,17 @@ func (enc *Encoder) encodeMap(b []byte, ctx encoderCtx, v reflect.Value) ([]byte
 
 	iter := v.MapRange()
 	for iter.Next() {
-		k := iter.Key().String()
 		v := iter.Value()
 
 		if isNil(v) {
 			continue
 		}
 
+		k, err := enc.keyToString(iter.Key())
+		if err != nil {
+			return nil, err
+		}
+
 		if willConvertToTableOrArrayTable(ctx, v) {
 			t.pushTable(k, v, emptyValueOptions)
 		} else {
@@ -688,6 +721,7 @@ func walkStruct(ctx encoderCtx, t *table, v reflect.Value) {
 		options := valueOptions{
 			multiline: opts.multiline,
 			omitempty: opts.omitempty,
+			commented: opts.commented,
 			comment:   fieldType.Tag.Get("comment"),
 		}
 
@@ -747,6 +781,7 @@ type tagOptions struct {
 	multiline bool
 	inline    bool
 	omitempty bool
+	commented bool
 }
 
 func parseTag(tag string) (string, tagOptions) {
@@ -774,6 +809,8 @@ func parseTag(tag string) (string, tagOptions) {
 			opts.inline = true
 		case "omitempty":
 			opts.omitempty = true
+		case "commented":
+			opts.commented = true
 		}
 	}
 
@@ -809,8 +846,10 @@ func (enc *Encoder) encodeTable(b []byte, ctx encoderCtx, t table) ([]byte, erro
 		hasNonEmptyKV = true
 
 		ctx.setKey(kv.Key)
+		ctx2 := ctx
+		ctx2.commented = kv.Options.commented || ctx2.commented
 
-		b, err = enc.encodeKv(b, ctx, kv.Options, kv.Value)
+		b, err = enc.encodeKv(b, ctx2, kv.Options, kv.Value)
 		if err != nil {
 			return nil, err
 		}
@@ -835,8 +874,10 @@ func (enc *Encoder) encodeTable(b []byte, ctx encoderCtx, t table) ([]byte, erro
 		ctx.setKey(table.Key)
 
 		ctx.options = table.Options
+		ctx2 := ctx
+		ctx2.commented = ctx2.commented || ctx.options.commented
 
-		b, err = enc.encode(b, ctx, table.Value)
+		b, err = enc.encode(b, ctx2, table.Value)
 		if err != nil {
 			return nil, err
 		}
@@ -914,7 +955,7 @@ func willConvertToTableOrArrayTable(ctx encoderCtx, v reflect.Value) bool {
 		return willConvertToTableOrArrayTable(ctx, v.Elem())
 	}
 
-	if t.Kind() == reflect.Slice {
+	if t.Kind() == reflect.Slice || t.Kind() == reflect.Array {
 		if v.Len() == 0 {
 			// An empty slice should be a kv = [].
 			return false
@@ -954,6 +995,9 @@ func (enc *Encoder) encodeSliceAsArrayTable(b []byte, ctx encoderCtx, v reflect.
 	ctx.shiftKey()
 
 	scratch := make([]byte, 0, 64)
+
+	scratch = enc.commented(ctx.commented, scratch)
+
 	scratch = append(scratch, "[["...)
 
 	for i, k := range ctx.parentKey {
@@ -969,6 +1013,10 @@ func (enc *Encoder) encodeSliceAsArrayTable(b []byte, ctx encoderCtx, v reflect.
 
 	b = enc.encodeComment(ctx.indent, ctx.options.comment, b)
 
+	if enc.indentTables {
+		ctx.indent++
+	}
+
 	for i := 0; i < v.Len(); i++ {
 		if i != 0 {
 			b = append(b, "\n"...)

marshaler_test.go

@@ -15,6 +15,21 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+type marshalTextKey struct {
+	A string
+	B string
+}
+
+func (k marshalTextKey) MarshalText() ([]byte, error) {
+	return []byte(k.A + "-" + k.B), nil
+}
+
+type marshalBadTextKey struct{}
+
+func (k marshalBadTextKey) MarshalText() ([]byte, error) {
+	return nil, fmt.Errorf("error")
+}
+
 func TestMarshal(t *testing.T) {
 	someInt := 42
 
@@ -97,6 +112,53 @@ also = 'that'
 a = 'test'
 `,
 		},
+		{
+			desc: `map with text key`,
+			v: map[marshalTextKey]string{
+				{A: "a", B: "1"}: "value 1",
+				{A: "a", B: "2"}: "value 2",
+				{A: "b", B: "1"}: "value 3",
+			},
+			expected: `a-1 = 'value 1'
+a-2 = 'value 2'
+b-1 = 'value 3'
+`,
+		},
+		{
+			desc: `table with text key`,
+			v: map[marshalTextKey]map[string]string{
+				{A: "a", B: "1"}: {"value": "foo"},
+			},
+			expected: `[a-1]
+value = 'foo'
+`,
+		},
+		{
+			desc: `map with ptr text key`,
+			v: map[*marshalTextKey]string{
+				{A: "a", B: "1"}: "value 1",
+				{A: "a", B: "2"}: "value 2",
+				{A: "b", B: "1"}: "value 3",
+			},
+			expected: `a-1 = 'value 1'
+a-2 = 'value 2'
+b-1 = 'value 3'
+`,
+		},
+		{
+			desc: `map with bad text key`,
+			v: map[marshalBadTextKey]string{
+				{}: "value 1",
+			},
+			err: true,
+		},
+		{
+			desc: `map with bad ptr text key`,
+			v: map[*marshalBadTextKey]string{
+				{}: "value 1",
+			},
+			err: true,
+		},
 		{
 			desc: "simple string array",
 			v: map[string][]string{
@@ -144,6 +206,45 @@ a = 'test'
 
 [[top]]
 'map2.1' = 'v2.1'
+`,
+		},
+		{
+			desc: "fixed size string array",
+			v: map[string][3]string{
+				"array": {"one", "two", "three"},
+			},
+			expected: `array = ['one', 'two', 'three']
+`,
+		},
+		{
+			desc: "fixed size nested string arrays",
+			v: map[string][2][2]string{
+				"array": {{"one", "two"}, {"three"}},
+			},
+			expected: `array = [['one', 'two'], ['three', '']]
+`,
+		},
+		{
+			desc: "mixed strings and fixed size nested string arrays",
+			v: map[string][]interface{}{
+				"array": {"a string", [2]string{"one", "two"}, "last"},
+			},
+			expected: `array = ['a string', ['one', 'two'], 'last']
+`,
+		},
+		{
+			desc: "fixed size array of maps",
+			v: map[string][2]map[string]string{
+				"ftop": {
+					{"map1.1": "v1.1"},
+					{"map2.1": "v2.1"},
+				},
+			},
+			expected: `[[ftop]]
+'map1.1' = 'v1.1'
+
+[[ftop]]
+'map2.1' = 'v2.1'
 `,
 		},
 		{
@@ -487,9 +588,14 @@ foo = 42
 		},
 		{
 			desc: "invalid map key",
-			v:    map[int]interface{}{},
+			v:    map[int]interface{}{1: "a"},
 			err:  true,
 		},
+		{
+			desc:     "invalid map key but empty",
+			v:        map[int]interface{}{},
+			expected: "",
+		},
 		{
 			desc: "unhandled type",
 			v: struct {
@@ -1095,6 +1201,45 @@ randomize = true
 	require.Equal(t, expected, buf.String())
 }
 
+func TestMarhsalIssue888(t *testing.T) {
+	type Thing struct {
+		FieldA string `comment:"my field A"`
+		FieldB string `comment:"my field B"`
+	}
+
+	type Cfg struct {
+		Custom []Thing `comment:"custom config"`
+	}
+
+	buf := new(bytes.Buffer)
+
+	config := Cfg{
+		Custom: []Thing{
+			{FieldA: "field a 1", FieldB: "field b 1"},
+			{FieldA: "field a 2", FieldB: "field b 2"},
+		},
+	}
+
+	encoder := toml.NewEncoder(buf).SetIndentTables(true)
+	encoder.Encode(config)
+
+	expected := `# custom config
+[[Custom]]
+  # my field A
+  FieldA = 'field a 1'
+  # my field B
+  FieldB = 'field b 1'
+
+[[Custom]]
+  # my field A
+  FieldA = 'field a 2'
+  # my field B
+  FieldB = 'field b 2'
+`
+
+	require.Equal(t, expected, buf.String())
+}
+
 func TestMarshalNestedAnonymousStructs(t *testing.T) {
 	type Embedded struct {
 		Value string `toml:"value" json:"value"`
@@ -1192,7 +1337,7 @@ func TestMarshalUint64Overflow(t *testing.T) {
 
 func TestIndentWithInlineTable(t *testing.T) {
 	x := map[string][]map[string]string{
-		"one": []map[string]string{
+		"one": {
 			{"0": "0"},
 			{"1": "1"},
 		},
@@ -1211,6 +1356,94 @@ func TestIndentWithInlineTable(t *testing.T) {
 	assert.Equal(t, expected, buf.String())
 }
 
+type C3 struct {
+	Value  int   `toml:",commented"`
+	Values []int `toml:",commented"`
+}
+
+type C2 struct {
+	Int       int64
+	String    string
+	ArrayInts []int
+	Structs   []C3
+}
+
+type C1 struct {
+	Int       int64  `toml:",commented"`
+	String    string `toml:",commented"`
+	ArrayInts []int  `toml:",commented"`
+	Structs   []C3   `toml:",commented"`
+}
+
+type Commented struct {
+	Int    int64  `toml:",commented"`
+	String string `toml:",commented"`
+
+	C1 C1
+	C2 C2 `toml:",commented"` // same as C1, but commented at top level
+}
+
+func TestMarshalCommented(t *testing.T) {
+	c := Commented{
+		Int:    42,
+		String: "root",
+
+		C1: C1{
+			Int:       11,
+			String:    "C1",
+			ArrayInts: []int{1, 2, 3},
+			Structs: []C3{
+				{Value: 100},
+				{Values: []int{4, 5, 6}},
+			},
+		},
+		C2: C2{
+			Int:       22,
+			String:    "C2",
+			ArrayInts: []int{1, 2, 3},
+			Structs: []C3{
+				{Value: 100},
+				{Values: []int{4, 5, 6}},
+			},
+		},
+	}
+
+	out, err := toml.Marshal(c)
+	require.NoError(t, err)
+
+	expected := `# Int = 42
+# String = 'root'
+
+[C1]
+# Int = 11
+# String = 'C1'
+# ArrayInts = [1, 2, 3]
+
+# [[C1.Structs]]
+# Value = 100
+# Values = []
+
+# [[C1.Structs]]
+# Value = 0
+# Values = [4, 5, 6]
+
+# [C2]
+# Int = 22
+# String = 'C2'
+# ArrayInts = [1, 2, 3]
+
+# [[C2.Structs]]
+# Value = 100
+# Values = []
+
+# [[C2.Structs]]
+# Value = 0
+# Values = [4, 5, 6]
+`
+
+	require.Equal(t, expected, string(out))
+}
+
 func ExampleMarshal() {
 	type MyConfig struct {
 		Version int
@@ -1235,3 +1468,161 @@ func ExampleMarshal() {
 	// Name = 'go-toml'
 	// Tags = ['go', 'toml']
 }
+
+// Example that uses the 'commented' field tag option to generate an example
+// configuration file that has commented out sections (example from
+// go-graphite/graphite-clickhouse).
+func ExampleMarshal_commented() {
+
+	type Common struct {
+		Listen               string        `toml:"listen"                     comment:"general listener"`
+		PprofListen          string        `toml:"pprof-listen"               comment:"listener to serve /debug/pprof requests. '-pprof' argument overrides it"`
+		MaxMetricsPerTarget  int           `toml:"max-metrics-per-target"     comment:"limit numbers of queried metrics per target in /render requests, 0 or negative = unlimited"`
+		MemoryReturnInterval time.Duration `toml:"memory-return-interval"     comment:"daemon will return the freed memory to the OS when it>0"`
+	}
+
+	type Costs struct {
+		Cost       *int           `toml:"cost"        comment:"default cost (for wildcarded equalence or matched with regex, or if no value cost set)"`
+		ValuesCost map[string]int `toml:"values-cost" comment:"cost with some value (for equalence without wildcards) (additional tuning, usually not needed)"`
+	}
+
+	type ClickHouse struct {
+		URL string `toml:"url" comment:"default url, see https://clickhouse.tech/docs/en/interfaces/http. Can be overwritten with query-params"`
+
+		RenderMaxQueries        int               `toml:"render-max-queries" comment:"Max queries to render queiries"`
+		RenderConcurrentQueries int               `toml:"render-concurrent-queries" comment:"Concurrent queries to render queiries"`
+		TaggedCosts             map[string]*Costs `toml:"tagged-costs,commented"`
+		TreeTable               string            `toml:"tree-table,commented"`
+		ReverseTreeTable        string            `toml:"reverse-tree-table,commented"`
+		DateTreeTable           string            `toml:"date-tree-table,commented"`
+		DateTreeTableVersion    int               `toml:"date-tree-table-version,commented"`
+		TreeTimeout             time.Duration     `toml:"tree-timeout,commented"`
+		TagTable                string            `toml:"tag-table,commented"`
+		ExtraPrefix             string            `toml:"extra-prefix"             comment:"add extra prefix (directory in graphite) for all metrics, w/o trailing dot"`
+		ConnectTimeout          time.Duration     `toml:"connect-timeout"          comment:"TCP connection timeout"`
+		DataTableLegacy         string            `toml:"data-table,commented"`
+		RollupConfLegacy        string            `toml:"rollup-conf,commented"`
+		MaxDataPoints           int               `toml:"max-data-points"          comment:"max points per metric when internal-aggregation=true"`
+		InternalAggregation     bool              `toml:"internal-aggregation"     comment:"ClickHouse-side aggregation, see doc/aggregation.md"`
+	}
+
+	type Tags struct {
+		Rules      string `toml:"rules"`
+		Date       string `toml:"date"`
+		ExtraWhere string `toml:"extra-where"`
+		InputFile  string `toml:"input-file"`
+		OutputFile string `toml:"output-file"`
+	}
+
+	type Config struct {
+		Common     Common     `toml:"common"`
+		ClickHouse ClickHouse `toml:"clickhouse"`
+		Tags       Tags       `toml:"tags,commented"`
+	}
+
+	cfg := &Config{
+		Common: Common{
+			Listen:               ":9090",
+			PprofListen:          "",
+			MaxMetricsPerTarget:  15000, // This is arbitrary value to protect CH from overload
+			MemoryReturnInterval: 0,
+		},
+		ClickHouse: ClickHouse{
+			URL:                 "http://localhost:8123?cancel_http_readonly_queries_on_client_close=1",
+			ExtraPrefix:         "",
+			ConnectTimeout:      time.Second,
+			DataTableLegacy:     "",
+			RollupConfLegacy:    "auto",
+			MaxDataPoints:       1048576,
+			InternalAggregation: true,
+		},
+		Tags: Tags{},
+	}
+
+	out, err := toml.Marshal(cfg)
+	if err != nil {
+		panic(err)
+	}
+	err = toml.Unmarshal(out, &cfg)
+	if err != nil {
+		panic(err)
+	}
+
+	fmt.Println(string(out))
+
+	// Output:
+	// [common]
+	// # general listener
+	// listen = ':9090'
+	// # listener to serve /debug/pprof requests. '-pprof' argument overrides it
+	// pprof-listen = ''
+	// # limit numbers of queried metrics per target in /render requests, 0 or negative = unlimited
+	// max-metrics-per-target = 15000
+	// # daemon will return the freed memory to the OS when it>0
+	// memory-return-interval = 0
+	//
+	// [clickhouse]
+	// # default url, see https://clickhouse.tech/docs/en/interfaces/http. Can be overwritten with query-params
+	// url = 'http://localhost:8123?cancel_http_readonly_queries_on_client_close=1'
+	// # Max queries to render queiries
+	// render-max-queries = 0
+	// # Concurrent queries to render queiries
+	// render-concurrent-queries = 0
+	// # tree-table = ''
+	// # reverse-tree-table = ''
+	// # date-tree-table = ''
+	// # date-tree-table-version = 0
+	// # tree-timeout = 0
+	// # tag-table = ''
+	// # add extra prefix (directory in graphite) for all metrics, w/o trailing dot
+	// extra-prefix = ''
+	// # TCP connection timeout
+	// connect-timeout = 1000000000
+	// # data-table = ''
+	// # rollup-conf = 'auto'
+	// # max points per metric when internal-aggregation=true
+	// max-data-points = 1048576
+	// # ClickHouse-side aggregation, see doc/aggregation.md
+	// internal-aggregation = true
+	//
+	// # [tags]
+	// # rules = ''
+	// # date = ''
+	// # extra-where = ''
+	// # input-file = ''
+	// # output-file = ''
+}
+
+func TestReadmeComments(t *testing.T) {
+	type TLS struct {
+		Cipher  string `toml:"cipher"`
+		Version string `toml:"version"`
+	}
+	type Config struct {
+		Host string `toml:"host" comment:"Host IP to connect to."`
+		Port int    `toml:"port" comment:"Port of the remote server."`
+		Tls  TLS    `toml:"TLS,commented" comment:"Encryption parameters (optional)"`
+	}
+	example := Config{
+		Host: "127.0.0.1",
+		Port: 4242,
+		Tls: TLS{
+			Cipher:  "AEAD-AES128-GCM-SHA256",
+			Version: "TLS 1.3",
+		},
+	}
+	out, err := toml.Marshal(example)
+	require.NoError(t, err)
+
+	expected := `# Host IP to connect to.
+host = '127.0.0.1'
+# Port of the remote server.
+port = 4242
+
+# Encryption parameters (optional)
+# [TLS]
+# cipher = 'AEAD-AES128-GCM-SHA256'
+# version = 'TLS 1.3'
+`
+	require.Equal(t, expected, string(out))
+}

ossfuzz/fuzz.go

@@ -0,0 +1,45 @@
+//go:build go1.18 || go1.19 || go1.20 || go1.21
+// +build go1.18 go1.19 go1.20 go1.21
+
+package ossfuzz
+
+import (
+	"fmt"
+	"reflect"
+	"strings"
+
+	"github.com/pelletier/go-toml/v2"
+)
+
+func FuzzToml(data []byte) int {
+	if len(data) >= 2048 {
+		return 0
+	}
+
+	if strings.Contains(string(data), "nan") {
+		return 0
+	}
+
+	var v interface{}
+	err := toml.Unmarshal(data, &v)
+	if err != nil {
+		return 0
+	}
+
+	encoded, err := toml.Marshal(v)
+	if err != nil {
+		panic(fmt.Sprintf("failed to marshal unmarshaled document: %s", err))
+	}
+
+	var v2 interface{}
+	err = toml.Unmarshal(encoded, &v2)
+	if err != nil {
+		panic(fmt.Sprintf("failed round trip: %s", err))
+	}
+
+	if !reflect.DeepEqual(v, v2) {
+		panic(fmt.Sprintf("not equal: %#+v %#+v", v, v2))
+	}
+
+	return 1
+}

unmarshaler.go

@@ -149,12 +149,16 @@ type errorContext struct {
 }
 
 func (d *decoder) typeMismatchError(toml string, target reflect.Type) error {
+	return fmt.Errorf("toml: %s", d.typeMismatchString(toml, target))
+}
+
+func (d *decoder) typeMismatchString(toml string, target reflect.Type) string {
 	if d.errorContext != nil && d.errorContext.Struct != nil {
 		ctx := d.errorContext
 		f := ctx.Struct.FieldByIndex(ctx.Field)
-		return fmt.Errorf("toml: cannot decode TOML %s into struct field %s.%s of type %s", toml, ctx.Struct, f.Name, f.Type)
+		return fmt.Sprintf("cannot decode TOML %s into struct field %s.%s of type %s", toml, ctx.Struct, f.Name, f.Type)
 	}
-	return fmt.Errorf("toml: cannot decode TOML %s into a Go value of type %s", toml, target)
+	return fmt.Sprintf("cannot decode TOML %s into a Go value of type %s", toml, target)
 }
 
 func (d *decoder) expr() *unstable.Node {
@@ -417,7 +421,10 @@ func (d *decoder) handleKeyPart(key unstable.Iterator, v reflect.Value, nextFn h
 		vt := v.Type()
 
 		// Create the key for the map element. Convert to key type.
-		mk := reflect.ValueOf(string(key.Node().Data)).Convert(vt.Key())
+		mk, err := d.keyFromData(vt.Key(), key.Node().Data)
+		if err != nil {
+			return reflect.Value{}, err
+		}
 
 		// If the map does not exist, create it.
 		if v.IsNil() {
@@ -960,7 +967,7 @@ func (d *decoder) unmarshalInteger(value *unstable.Node, v reflect.Value) error
 	case reflect.Interface:
 		r = reflect.ValueOf(i)
 	default:
-		return d.typeMismatchError("integer", v.Type())
+		return unstable.NewParserError(d.p.Raw(value.Raw), d.typeMismatchString("integer", v.Type()))
 	}
 
 	if !r.Type().AssignableTo(v.Type()) {
@@ -979,7 +986,7 @@ func (d *decoder) unmarshalString(value *unstable.Node, v reflect.Value) error {
 	case reflect.Interface:
 		v.Set(reflect.ValueOf(string(value.Data)))
 	default:
-		return unstable.NewParserError(d.p.Raw(value.Raw), "cannot store TOML string into a Go %s", v.Kind())
+		return unstable.NewParserError(d.p.Raw(value.Raw), d.typeMismatchString("string", v.Type()))
 	}
 
 	return nil
@@ -1009,6 +1016,31 @@ func (d *decoder) handleKeyValueInner(key unstable.Iterator, value *unstable.Nod
 	return reflect.Value{}, d.handleValue(value, v)
 }
 
+func (d *decoder) keyFromData(keyType reflect.Type, data []byte) (reflect.Value, error) {
+	switch {
+	case stringType.AssignableTo(keyType):
+		return reflect.ValueOf(string(data)), nil
+
+	case stringType.ConvertibleTo(keyType):
+		return reflect.ValueOf(string(data)).Convert(keyType), nil
+
+	case keyType.Implements(textUnmarshalerType):
+		mk := reflect.New(keyType.Elem())
+		if err := mk.Interface().(encoding.TextUnmarshaler).UnmarshalText(data); err != nil {
+			return reflect.Value{}, fmt.Errorf("toml: error unmarshalling key type %s from text: %w", stringType, err)
+		}
+		return mk, nil
+
+	case reflect.PtrTo(keyType).Implements(textUnmarshalerType):
+		mk := reflect.New(keyType)
+		if err := mk.Interface().(encoding.TextUnmarshaler).UnmarshalText(data); err != nil {
+			return reflect.Value{}, fmt.Errorf("toml: error unmarshalling key type %s from text: %w", stringType, err)
+		}
+		return mk.Elem(), nil
+	}
+	return reflect.Value{}, fmt.Errorf("toml: cannot convert map key of type %s to expected type %s", stringType, keyType)
+}
+
 func (d *decoder) handleKeyValuePart(key unstable.Iterator, value *unstable.Node, v reflect.Value) (reflect.Value, error) {
 	// contains the replacement for v
 	var rv reflect.Value
@@ -1019,16 +1051,9 @@ func (d *decoder) handleKeyValuePart(key unstable.Iterator, value *unstable.Node
 	case reflect.Map:
 		vt := v.Type()
 
-		mk := reflect.ValueOf(string(key.Node().Data))
+		mk, err := d.keyFromData(vt.Key(), key.Node().Data)
-		mkt := stringType
+		if err != nil {
-
+			return reflect.Value{}, err
-		keyType := vt.Key()
-		if !mkt.AssignableTo(keyType) {
-			if !mkt.ConvertibleTo(keyType) {
-				return reflect.Value{}, fmt.Errorf("toml: cannot convert map key of type %s to expected type %s", mkt, keyType)
-			}
-
-			mk = mk.Convert(keyType)
 		}
 
 		// If the map does not exist, create it.
@@ -1071,6 +1096,19 @@ func (d *decoder) handleKeyValuePart(key unstable.Iterator, value *unstable.Node
 		d.errorContext.Field = path
 
 		f := fieldByIndex(v, path)
+
+		if !f.CanSet() {
+			// If the field is not settable, need to take a slower path and make a copy of
+			// the struct itself to a new location.
+			nvp := reflect.New(v.Type())
+			nvp.Elem().Set(v)
+			v = nvp.Elem()
+			_, err := d.handleKeyValuePart(key, value, v)
+			if err != nil {
+				return reflect.Value{}, err
+			}
+			return nvp.Elem(), nil
+		}
 		x, err := d.handleKeyValueInner(key, value, f)
 		if err != nil {
 			return reflect.Value{}, err
@@ -1136,10 +1174,10 @@ func initAndDereferencePointer(v reflect.Value) reflect.Value {
 
 // Same as reflect.Value.FieldByIndex, but creates pointers if needed.
 func fieldByIndex(v reflect.Value, path []int) reflect.Value {
-	for i, x := range path {
+	for _, x := range path {
 		v = v.Field(x)
 
-		if i < len(path)-1 && v.Kind() == reflect.Ptr {
+		if v.Kind() == reflect.Ptr {
 			if v.IsNil() {
 				v.Set(reflect.New(v.Type().Elem()))
 			}

unmarshaler_test.go

@@ -16,6 +16,27 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+type unmarshalTextKey struct {
+	A string
+	B string
+}
+
+func (k *unmarshalTextKey) UnmarshalText(text []byte) error {
+	parts := strings.Split(string(text), "-")
+	if len(parts) != 2 {
+		return fmt.Errorf("invalid text key: %s", text)
+	}
+	k.A = parts[0]
+	k.B = parts[1]
+	return nil
+}
+
+type unmarshalBadTextKey struct{}
+
+func (k *unmarshalBadTextKey) UnmarshalText(text []byte) error {
+	return fmt.Errorf("error")
+}
+
 func ExampleDecoder_DisallowUnknownFields() {
 	type S struct {
 		Key1 string
@@ -69,7 +90,6 @@ func ExampleUnmarshal() {
 	fmt.Println("version:", cfg.Version)
 	fmt.Println("name:", cfg.Name)
 	fmt.Println("tags:", cfg.Tags)
-
 	// Output:
 	// version: 2
 	// name: go-toml
@@ -220,6 +240,11 @@ func TestUnmarshal_Floats(t *testing.T) {
 			input:    `0E0`,
 			expected: 0.0,
 		},
+		{
+			desc:     "float zero without decimals",
+			input:    `0`,
+			expected: 0.0,
+		},
 		{
 			desc:     "float fractional with exponent",
 			input:    `6.626e-34`,
@@ -315,6 +340,7 @@ func TestUnmarshal(t *testing.T) {
 		target   interface{}
 		expected interface{}
 		err      bool
+		assert   func(t *testing.T, test test)
 	}
 	examples := []struct {
 		skip  bool
@@ -350,6 +376,96 @@ func TestUnmarshal(t *testing.T) {
 				}
 			},
 		},
+		{
+			desc:  "kv text key",
+			input: `a-1 = "foo"`,
+			gen: func() test {
+				type doc = map[unmarshalTextKey]string
+
+				return test{
+					target:   &doc{},
+					expected: &doc{{A: "a", B: "1"}: "foo"},
+				}
+			},
+		},
+		{
+			desc: "table text key",
+			input: `["a-1"]
+foo = "bar"`,
+			gen: func() test {
+				type doc = map[unmarshalTextKey]map[string]string
+
+				return test{
+					target:   &doc{},
+					expected: &doc{{A: "a", B: "1"}: map[string]string{"foo": "bar"}},
+				}
+			},
+		},
+		{
+			desc:  "kv ptr text key",
+			input: `a-1 = "foo"`,
+			gen: func() test {
+				type doc = map[*unmarshalTextKey]string
+
+				return test{
+					target:   &doc{},
+					expected: &doc{{A: "a", B: "1"}: "foo"},
+					assert: func(t *testing.T, test test) {
+						// Despite the documentation:
+						//     Pointer variable equality is determined based on the equality of the
+						// 		 referenced values (as opposed to the memory addresses).
+						// assert.Equal does not work properly with maps with pointer keys
+						// https://github.com/stretchr/testify/issues/1143
+						expected := make(map[unmarshalTextKey]string)
+						for k, v := range *(test.expected.(*doc)) {
+							expected[*k] = v
+						}
+						got := make(map[unmarshalTextKey]string)
+						for k, v := range *(test.target.(*doc)) {
+							got[*k] = v
+						}
+						assert.Equal(t, expected, got)
+					},
+				}
+			},
+		},
+		{
+			desc:  "kv bad text key",
+			input: `a-1 = "foo"`,
+			gen: func() test {
+				type doc = map[unmarshalBadTextKey]string
+
+				return test{
+					target: &doc{},
+					err:    true,
+				}
+			},
+		},
+		{
+			desc:  "kv bad ptr text key",
+			input: `a-1 = "foo"`,
+			gen: func() test {
+				type doc = map[*unmarshalBadTextKey]string
+
+				return test{
+					target: &doc{},
+					err:    true,
+				}
+			},
+		},
+		{
+			desc: "table bad text key",
+			input: `["a-1"]
+foo = "bar"`,
+			gen: func() test {
+				type doc = map[unmarshalBadTextKey]map[string]string
+
+				return test{
+					target: &doc{},
+					err:    true,
+				}
+			},
+		},
 		{
 			desc:  "time.time with negative zone",
 			input: `a = 1979-05-27T00:32:00-07:00 `, // space intentional
@@ -1166,6 +1282,64 @@ B = "data"`,
 				}
 			},
 		},
+		{
+			desc: "array table into maps with pointer on last key",
+			input: `[[foo]]
+			bar = "hello"`,
+			gen: func() test {
+				type doc struct {
+					Foo **[]interface{}
+				}
+				x := &[]interface{}{
+					map[string]interface{}{
+						"bar": "hello",
+					},
+				}
+				return test{
+					target: &doc{},
+					expected: &doc{
+						Foo: &x,
+					},
+				}
+			},
+		},
+		{
+			desc: "array table into maps with pointer on intermediate key",
+			input: `[[foo.foo2]]
+			bar = "hello"`,
+			gen: func() test {
+				type doc struct {
+					Foo **map[string]interface{}
+				}
+				x := &map[string]interface{}{
+					"foo2": []interface{}{
+						map[string]interface{}{
+							"bar": "hello",
+						},
+					},
+				}
+				return test{
+					target: &doc{},
+					expected: &doc{
+						Foo: &x,
+					},
+				}
+			},
+		},
+		{
+			desc: "array table into maps with pointer on last key with invalid leaf type",
+			input: `[[foo]]
+			bar = "hello"`,
+			gen: func() test {
+				type doc struct {
+					Foo **[]map[string]int
+				}
+				return test{
+					target: &doc{},
+					err:    true,
+				}
+			},
+		},
 		{
 			desc:  "unexported struct fields are ignored",
 			input: `foo = "bar"`,
@@ -1521,6 +1695,16 @@ B = "data"`,
 				}
 			},
 		},
+		{
+			desc:  "empty map into map with invalid key type",
+			input: ``,
+			gen: func() test {
+				return test{
+					target:   &map[int]string{},
+					expected: &map[int]string{},
+				}
+			},
+		},
 		{
 			desc:  "into map with convertible key type",
 			input: `A = "hello"`,
@@ -1777,7 +1961,11 @@ B = "data"`,
 				require.Error(t, err)
 			} else {
 				require.NoError(t, err)
-				assert.Equal(t, test.expected, test.target)
+				if test.assert != nil {
+					test.assert(t, test)
+				} else {
+					assert.Equal(t, test.expected, test.target)
+				}
 			}
 		})
 	}
@@ -1860,6 +2048,68 @@ func TestUnmarshalErrors(t *testing.T) {
 	require.Equal(t, "toml: cannot decode TOML integer into struct field toml_test.mystruct.Bar of type string", err.Error())
 }
 
+func TestUnmarshalStringInvalidStructField(t *testing.T) {
+	type Server struct {
+		Path string
+		Port int
+	}
+
+	type Cfg struct {
+		Server Server
+	}
+
+	var cfg Cfg
+
+	data := `[server]
+path = "/my/path"
+port = "bad"
+`
+
+	file := strings.NewReader(data)
+	err := toml.NewDecoder(file).Decode(&cfg)
+	require.Error(t, err)
+
+	x := err.(*toml.DecodeError)
+	require.Equal(t, "toml: cannot decode TOML string into struct field toml_test.Server.Port of type int", x.Error())
+	expected := `1| [server]
+2| path = "/my/path"
+3| port = "bad"
+ |        ~~~~~ cannot decode TOML string into struct field toml_test.Server.Port of type int`
+
+	require.Equal(t, expected, x.String())
+}
+
+func TestUnmarshalIntegerInvalidStructField(t *testing.T) {
+	type Server struct {
+		Path string
+		Port int
+	}
+
+	type Cfg struct {
+		Server Server
+	}
+
+	var cfg Cfg
+
+	data := `[server]
+path = 100
+port = 50
+`
+
+	file := strings.NewReader(data)
+	err := toml.NewDecoder(file).Decode(&cfg)
+	require.Error(t, err)
+
+	x := err.(*toml.DecodeError)
+	require.Equal(t, "toml: cannot decode TOML integer into struct field toml_test.Server.Path of type string", x.Error())
+	expected := `1| [server]
+2| path = 100
+ |        ~~~ cannot decode TOML integer into struct field toml_test.Server.Path of type string
+3| port = 50`
+
+	require.Equal(t, expected, x.String())
+}
+
 func TestUnmarshalInvalidTarget(t *testing.T) {
 	x := "foo"
 	err := toml.Unmarshal([]byte{}, x)
@@ -2487,6 +2737,70 @@ func TestIssue851(t *testing.T) {
 	require.Equal(t, map[string]string{"a": "1", "b": "2"}, target.Params)
 }
 
+func TestIssue866(t *testing.T) {
+	type Pipeline struct {
+		Mapping map[string]struct {
+			Req [][]string `toml:"req"`
+			Res [][]string `toml:"res"`
+		} `toml:"mapping"`
+	}
+
+	type Pipelines struct {
+		PipelineMapping map[string]*Pipeline `toml:"pipelines"`
+	}
+
+	var badToml = `
+[pipelines.register]
+mapping.inst.req = [
+    ["param1", "value1"],
+]
+mapping.inst.res = [
+    ["param2", "value2"],
+]
+`
+
+	pipelines := new(Pipelines)
+	if err := toml.NewDecoder(bytes.NewBufferString(badToml)).DisallowUnknownFields().Decode(pipelines); err != nil {
+		t.Fatal(err)
+	}
+	if pipelines.PipelineMapping["register"].Mapping["inst"].Req[0][0] != "param1" {
+		t.Fatal("unmarshal failed with mismatch value")
+	}
+
+	var goodTooToml = `
+[pipelines.register]
+mapping.inst.req = [
+    ["param1", "value1"],
+]
+`
+
+	pipelines = new(Pipelines)
+	if err := toml.NewDecoder(bytes.NewBufferString(goodTooToml)).DisallowUnknownFields().Decode(pipelines); err != nil {
+		t.Fatal(err)
+	}
+	if pipelines.PipelineMapping["register"].Mapping["inst"].Req[0][0] != "param1" {
+		t.Fatal("unmarshal failed with mismatch value")
+	}
+
+	var goodToml = `
+[pipelines.register.mapping.inst]
+req = [
+    ["param1", "value1"],
+]
+res = [
+    ["param2", "value2"],
+]
+`
+
+	pipelines = new(Pipelines)
+	if err := toml.NewDecoder(bytes.NewBufferString(goodToml)).DisallowUnknownFields().Decode(pipelines); err != nil {
+		t.Fatal(err)
+	}
+	if pipelines.PipelineMapping["register"].Mapping["inst"].Req[0][0] != "param1" {
+		t.Fatal("unmarshal failed with mismatch value")
+	}
+}
+
 func TestUnmarshalDecodeErrors(t *testing.T) {
 	examples := []struct {
 		desc string
@@ -3322,3 +3636,47 @@ func TestUnmarshalEmbedNonString(t *testing.T) {
 	require.NoError(t, err)
 	require.Nil(t, d.Foo)
 }
+
+func TestUnmarshal_Nil(t *testing.T) {
+	type Foo struct {
+		Foo *Foo `toml:"foo,omitempty"`
+		Bar *Foo `toml:"bar,omitempty"`
+	}
+
+	examples := []struct {
+		desc     string
+		input    string
+		expected string
+		err      bool
+	}{
+		{
+			desc:     "empty",
+			input:    ``,
+			expected: ``,
+		},
+		{
+			desc: "simplest",
+			input: `
+            [foo]
+            [foo.foo]
+            `,
+			expected: "[foo]\n[foo.foo]\n",
+		},
+	}
+
+	for _, ex := range examples {
+		e := ex
+		t.Run(e.desc, func(t *testing.T) {
+			foo := Foo{}
+			err := toml.Unmarshal([]byte(e.input), &foo)
+			if e.err {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+				j, err := toml.Marshal(foo)
+				require.NoError(t, err)
+				assert.Equal(t, e.expected, string(j))
+			}
+		})
+	}
+}

unstable/parser.go

@@ -49,8 +49,6 @@ func NewParserError(highlight []byte, format string, args ...interface{}) error
 // For performance reasons, go-toml doesn't make a copy of the input bytes to
 // the parser. Make sure to copy all the bytes you need to outlive the slice
 // given to the parser.
-//
-// The parser doesn't provide nodes for comments yet, nor for whitespace.
 type Parser struct {
 	data    []byte
 	builder builder
@@ -58,6 +56,8 @@ type Parser struct {
 	left    []byte
 	err     error
 	first   bool
+
+	KeepComments bool
 }
 
 // Data returns the slice provided to the last call to Reset.
@@ -142,6 +142,44 @@ func (p *Parser) Error() error {
 	return p.err
 }
 
+// Position describes a position in the input.
+type Position struct {
+	// Number of bytes from the beginning of the input.
+	Offset int
+	// Line number, starting at 1.
+	Line int
+	// Column number, starting at 1.
+	Column int
+}
+
+// Shape describes the position of a range in the input.
+type Shape struct {
+	Start Position
+	End   Position
+}
+
+func (p *Parser) position(b []byte) Position {
+	offset := danger.SubsliceOffset(p.data, b)
+
+	lead := p.data[:offset]
+
+	return Position{
+		Offset: offset,
+		Line:   bytes.Count(lead, []byte{'\n'}) + 1,
+		Column: len(lead) - bytes.LastIndex(lead, []byte{'\n'}),
+	}
+}
+
+// Shape returns the shape of the given range in the input.  Will
+// panic if the range is not a subslice of the input.
+func (p *Parser) Shape(r Range) Shape {
+	raw := p.Raw(r)
+	return Shape{
+		Start: p.position(raw),
+		End:   p.position(raw[r.Length:]),
+	}
+}
+
 func (p *Parser) parseNewline(b []byte) ([]byte, error) {
 	if b[0] == '\n' {
 		return b[1:], nil
@@ -155,6 +193,19 @@ func (p *Parser) parseNewline(b []byte) ([]byte, error) {
 	return nil, NewParserError(b[0:1], "expected newline but got %#U", b[0])
 }
 
+func (p *Parser) parseComment(b []byte) (reference, []byte, error) {
+	ref := invalidReference
+	data, rest, err := scanComment(b)
+	if p.KeepComments && err == nil {
+		ref = p.builder.Push(Node{
+			Kind: Comment,
+			Raw:  p.Range(data),
+			Data: data,
+		})
+	}
+	return ref, rest, err
+}
+
 func (p *Parser) parseExpression(b []byte) (reference, []byte, error) {
 	// expression =  ws [ comment ]
 	// expression =/ ws keyval ws [ comment ]
@@ -168,7 +219,7 @@ func (p *Parser) parseExpression(b []byte) (reference, []byte, error) {
 	}
 
 	if b[0] == '#' {
-		_, rest, err := scanComment(b)
+		ref, rest, err := p.parseComment(b)
 		return ref, rest, err
 	}
 
@@ -190,7 +241,10 @@ func (p *Parser) parseExpression(b []byte) (reference, []byte, error) {
 	b = p.parseWhitespace(b)
 
 	if len(b) > 0 && b[0] == '#' {
-		_, rest, err := scanComment(b)
+		cref, rest, err := p.parseComment(b)
+		if cref != invalidReference {
+			p.builder.Chain(ref, cref)
+		}
 		return ref, rest, err
 	}
 
@@ -471,17 +525,33 @@ func (p *Parser) parseValArray(b []byte) (reference, []byte, error) {
 		Kind: Array,
 	})
 
+	// First indicates whether the parser is looking for the first element
+	// (non-comment) of the array.
 	first := true
 
-	var lastChild reference
+	lastChild := invalidReference
+
+	addChild := func(valueRef reference) {
+		if lastChild == invalidReference {
+			p.builder.AttachChild(parent, valueRef)
+		} else {
+			p.builder.Chain(lastChild, valueRef)
+		}
+		lastChild = valueRef
+	}
 
 	var err error
 	for len(b) > 0 {
-		b, err = p.parseOptionalWhitespaceCommentNewline(b)
+		cref := invalidReference
+		cref, b, err = p.parseOptionalWhitespaceCommentNewline(b)
 		if err != nil {
 			return parent, nil, err
 		}
 
+		if cref != invalidReference {
+			addChild(cref)
+		}
+
 		if len(b) == 0 {
 			return parent, nil, NewParserError(arrayStart[:1], "array is incomplete")
 		}
@@ -496,10 +566,13 @@ func (p *Parser) parseValArray(b []byte) (reference, []byte, error) {
 			}
 			b = b[1:]
 
-			b, err = p.parseOptionalWhitespaceCommentNewline(b)
+			cref, b, err = p.parseOptionalWhitespaceCommentNewline(b)
 			if err != nil {
 				return parent, nil, err
 			}
+			if cref != invalidReference {
+				addChild(cref)
+			}
 		} else if !first {
 			return parent, nil, NewParserError(b[0:1], "array elements must be separated by commas")
 		}
@@ -515,17 +588,16 @@ func (p *Parser) parseValArray(b []byte) (reference, []byte, error) {
 			return parent, nil, err
 		}
 
-		if first {
+		addChild(valueRef)
-			p.builder.AttachChild(parent, valueRef)
-		} else {
-			p.builder.Chain(lastChild, valueRef)
-		}
-		lastChild = valueRef
 
-		b, err = p.parseOptionalWhitespaceCommentNewline(b)
+		cref, b, err = p.parseOptionalWhitespaceCommentNewline(b)
 		if err != nil {
 			return parent, nil, err
 		}
+		if cref != invalidReference {
+			addChild(cref)
+		}
+
 		first = false
 	}
 
@@ -534,15 +606,34 @@ func (p *Parser) parseValArray(b []byte) (reference, []byte, error) {
 	return parent, rest, err
 }
 
-func (p *Parser) parseOptionalWhitespaceCommentNewline(b []byte) ([]byte, error) {
+func (p *Parser) parseOptionalWhitespaceCommentNewline(b []byte) (reference, []byte, error) {
+	rootCommentRef := invalidReference
+	latestCommentRef := invalidReference
+
+	addComment := func(ref reference) {
+		if rootCommentRef == invalidReference {
+			rootCommentRef = ref
+		} else if latestCommentRef == invalidReference {
+			p.builder.AttachChild(rootCommentRef, ref)
+			latestCommentRef = ref
+		} else {
+			p.builder.Chain(latestCommentRef, ref)
+			latestCommentRef = ref
+		}
+	}
+
 	for len(b) > 0 {
 		var err error
 		b = p.parseWhitespace(b)
 
 		if len(b) > 0 && b[0] == '#' {
-			_, b, err = scanComment(b)
+			var ref reference
+			ref, b, err = p.parseComment(b)
 			if err != nil {
-				return nil, err
+				return invalidReference, nil, err
+			}
+			if ref != invalidReference {
+				addComment(ref)
 			}
 		}
 
@@ -553,14 +644,14 @@ func (p *Parser) parseOptionalWhitespaceCommentNewline(b []byte) ([]byte, error)
 		if b[0] == '\n' || b[0] == '\r' {
 			b, err = p.parseNewline(b)
 			if err != nil {
-				return nil, err
+				return invalidReference, nil, err
 			}
 		} else {
 			break
 		}
 	}
 
-	return b, nil
+	return rootCommentRef, b, nil
 }
 
 func (p *Parser) parseMultilineLiteralString(b []byte) ([]byte, []byte, []byte, error) {
@@ -922,6 +1013,7 @@ func (p *Parser) parseIntOrFloatOrDateTime(b []byte) (reference, []byte, error)
 		return p.builder.Push(Node{
 			Kind: Float,
 			Data: b[:3],
+			Raw:  p.Range(b[:3]),
 		}), b[3:], nil
 	case 'n':
 		if !scanFollowsNan(b) {
@@ -931,6 +1023,7 @@ func (p *Parser) parseIntOrFloatOrDateTime(b []byte) (reference, []byte, error)
 		return p.builder.Push(Node{
 			Kind: Float,
 			Data: b[:3],
+			Raw:  p.Range(b[:3]),
 		}), b[3:], nil
 	case '+', '-':
 		return p.scanIntOrFloat(b)
@@ -1055,6 +1148,7 @@ func (p *Parser) scanIntOrFloat(b []byte) (reference, []byte, error) {
 		return p.builder.Push(Node{
 			Kind: Integer,
 			Data: b[:i],
+			Raw:  p.Range(b[:i]),
 		}), b[i:], nil
 	}
 
@@ -1078,6 +1172,7 @@ func (p *Parser) scanIntOrFloat(b []byte) (reference, []byte, error) {
 				return p.builder.Push(Node{
 					Kind: Float,
 					Data: b[:i+3],
+					Raw:  p.Range(b[:i+3]),
 				}), b[i+3:], nil
 			}
 
@@ -1089,6 +1184,7 @@ func (p *Parser) scanIntOrFloat(b []byte) (reference, []byte, error) {
 				return p.builder.Push(Node{
 					Kind: Float,
 					Data: b[:i+3],
+					Raw:  p.Range(b[:i+3]),
 				}), b[i+3:], nil
 			}
 
@@ -1111,6 +1207,7 @@ func (p *Parser) scanIntOrFloat(b []byte) (reference, []byte, error) {
 	return p.builder.Push(Node{
 		Kind: kind,
 		Data: b[:i],
+		Raw:  p.Range(b[:i]),
 	}), b[i:], nil
 }
 

unstable/parser_test.go

@@ -448,6 +448,163 @@ func TestParser_AST_DateTimes(t *testing.T) {
 	}
 }
 
+// This example demonstrates how to parse a TOML document and preserving
+// comments.  Comments are stored in the AST as Comment nodes. This example
+// displays the structure of the full AST generated by the parser using the
+// following structure:
+//
+//  1. Each root-level expression is separated by three dashes.
+//  2. Bytes associated to a node are displayed in square brackets.
+//  3. Siblings have the same indentation.
+//  4. Children of a node are indented one level.
+func ExampleParser_comments() {
+	doc := `# Top of the document comment.
+# Optional, any amount of lines.
+
+# Above table.
+[table] # Next to table.
+# Above simple value.
+key = "value" # Next to simple value.
+# Below simple value.
+
+# Some comment alone.
+
+# Multiple comments, on multiple lines.
+
+# Above inline table.
+name = { first = "Tom", last = "Preston-Werner" } # Next to inline table.
+# Below inline table.
+
+# Above array.
+array = [ 1, 2, 3 ] # Next to one-line array.
+# Below array.
+
+# Above multi-line array.
+key5 = [ # Next to start of inline array.
+  # Second line before array content.
+  1, # Next to first element.
+  # After first element.
+  # Before second element.
+  2,
+  3, # Next to last element
+  # After last element.
+] # Next to end of array.
+# Below multi-line array.
+
+# Before array table.
+[[products]] # Next to array table.
+# After array table.
+`
+
+	var printGeneric func(*Parser, int, *Node)
+	printGeneric = func(p *Parser, indent int, e *Node) {
+		if e == nil {
+			return
+		}
+		s := p.Shape(e.Raw)
+		x := fmt.Sprintf("%d:%d->%d:%d (%d->%d)", s.Start.Line, s.Start.Column, s.End.Line, s.End.Column, s.Start.Offset, s.End.Offset)
+		fmt.Printf("%-25s | %s%s [%s]\n", x, strings.Repeat("  ", indent), e.Kind, e.Data)
+		printGeneric(p, indent+1, e.Child())
+		printGeneric(p, indent, e.Next())
+	}
+
+	printTree := func(p *Parser) {
+		for p.NextExpression() {
+			e := p.Expression()
+			fmt.Println("---")
+			printGeneric(p, 0, e)
+		}
+		if err := p.Error(); err != nil {
+			panic(err)
+		}
+	}
+
+	p := &Parser{
+		KeepComments: true,
+	}
+	p.Reset([]byte(doc))
+	printTree(p)
+
+	// Output:
+	// ---
+	// 1:1->1:31 (0->30)         | Comment [# Top of the document comment.]
+	// ---
+	// 2:1->2:33 (31->63)        | Comment [# Optional, any amount of lines.]
+	// ---
+	// 4:1->4:15 (65->79)        | Comment [# Above table.]
+	// ---
+	// 1:1->1:1 (0->0)           | Table []
+	// 5:2->5:7 (81->86)         |   Key [table]
+	// 5:9->5:25 (88->104)       | Comment [# Next to table.]
+	// ---
+	// 6:1->6:22 (105->126)      | Comment [# Above simple value.]
+	// ---
+	// 1:1->1:1 (0->0)           | KeyValue []
+	// 7:7->7:14 (133->140)      |   String [value]
+	// 7:1->7:4 (127->130)       |   Key [key]
+	// 7:15->7:38 (141->164)     | Comment [# Next to simple value.]
+	// ---
+	// 8:1->8:22 (165->186)      | Comment [# Below simple value.]
+	// ---
+	// 10:1->10:22 (188->209)    | Comment [# Some comment alone.]
+	// ---
+	// 12:1->12:40 (211->250)    | Comment [# Multiple comments, on multiple lines.]
+	// ---
+	// 14:1->14:22 (252->273)    | Comment [# Above inline table.]
+	// ---
+	// 1:1->1:1 (0->0)           | KeyValue []
+	// 15:8->15:9 (281->282)     |   InlineTable []
+	// 1:1->1:1 (0->0)           |     KeyValue []
+	// 15:18->15:23 (291->296)   |       String [Tom]
+	// 15:10->15:15 (283->288)   |       Key [first]
+	// 1:1->1:1 (0->0)           |     KeyValue []
+	// 15:32->15:48 (305->321)   |       String [Preston-Werner]
+	// 15:25->15:29 (298->302)   |       Key [last]
+	// 15:1->15:5 (274->278)     |   Key [name]
+	// 15:51->15:74 (324->347)   | Comment [# Next to inline table.]
+	// ---
+	// 16:1->16:22 (348->369)    | Comment [# Below inline table.]
+	// ---
+	// 18:1->18:15 (371->385)    | Comment [# Above array.]
+	// ---
+	// 1:1->1:1 (0->0)           | KeyValue []
+	// 1:1->1:1 (0->0)           |   Array []
+	// 19:11->19:12 (396->397)   |     Integer [1]
+	// 19:14->19:15 (399->400)   |     Integer [2]
+	// 19:17->19:18 (402->403)   |     Integer [3]
+	// 19:1->19:6 (386->391)     |   Key [array]
+	// 19:21->19:46 (406->431)   | Comment [# Next to one-line array.]
+	// ---
+	// 20:1->20:15 (432->446)    | Comment [# Below array.]
+	// ---
+	// 22:1->22:26 (448->473)    | Comment [# Above multi-line array.]
+	// ---
+	// 1:1->1:1 (0->0)           | KeyValue []
+	// 1:1->1:1 (0->0)           |   Array []
+	// 23:10->23:42 (483->515)   |     Comment [# Next to start of inline array.]
+	// 24:3->24:38 (518->553)    |       Comment [# Second line before array content.]
+	// 25:3->25:4 (556->557)     |     Integer [1]
+	// 25:6->25:30 (559->583)    |     Comment [# Next to first element.]
+	// 26:3->26:25 (586->608)    |       Comment [# After first element.]
+	// 27:3->27:27 (611->635)    |       Comment [# Before second element.]
+	// 28:3->28:4 (638->639)     |     Integer [2]
+	// 29:3->29:4 (643->644)     |     Integer [3]
+	// 29:6->29:28 (646->668)    |     Comment [# Next to last element]
+	// 30:3->30:24 (671->692)    |       Comment [# After last element.]
+	// 23:1->23:5 (474->478)     |   Key [key5]
+	// 31:3->31:26 (695->718)    | Comment [# Next to end of array.]
+	// ---
+	// 32:1->32:26 (719->744)    | Comment [# Below multi-line array.]
+	// ---
+	// 34:1->34:22 (746->767)    | Comment [# Before array table.]
+	// ---
+	// 1:1->1:1 (0->0)           | ArrayTable []
+	// 35:3->35:11 (770->778)    |   Key [products]
+	// 35:14->35:36 (781->803)   | Comment [# Next to array table.]
+	// ---
+	// 36:1->36:21 (804->824)    | Comment [# After array table.]
+}
+
 func ExampleParser() {
 	doc := `
 	hello = "world"

unstable/scanner.go

@@ -151,7 +151,6 @@ func scanWhitespace(b []byte) ([]byte, []byte) {
 	return b, b[len(b):]
 }
 
-//nolint:unparam
 func scanComment(b []byte) ([]byte, []byte, error) {
 	// comment-start-symbol = %x23 ; #
 	// non-ascii = %x80-D7FF / %xE000-10FFFF